challengedchip

It seems like a stretch to argue that the model isn't "a work based on" GPL code when that GPL code is an input to a deterministic algorithm from which the model is produced. So, my bet is on point #1.

The only ambiguity as far as I can tell is GPL covers "source code", "machine-readable Corresponding Source", and "object code form", and it's not explicit whether vector-fields count as any of those things. I doubt anyone would seriously argue that zipping and then un-zipping some GPL source code means you don't need to respect the original license. LLMs are different in that they're lossy compared to the zip format - does the nature of this lossiness invalidate the intent of the GPL's original language? I doubt it.

It could be as clear-cut as you've just made it: "a human". An LLM is not a human.

You could get into the semantics of "learning" - does JPEG encoding count as the computer "learning" how to reproduce the original image? But trying to create some metric for why LLMs "learn" and JPEG doesn't "learn" on the basis of the algorithms is a philosophical endeavor. Copyright is more about practicality - about realized externalities - than it is about philosophy. That's why selling cars and selling guns are regulated differently, despite the fact that you could reduce both to "metal mechanical machines that kill" by rhetorical argument.

Even from a strictly legal perspective, it actually is fairly clear-cut. The answer to "what if I (a human) read GPL code and then reuse the knowledge gained from it..." comes down to a few straightforward properties of the license. GPL doesn't cover "reduced to practice" as many corporate contracts do, so terms covering "the knowledge gained" are lenient. GPL covers "verbatim" copies which is what LLMs are doing, that's as clear cut as it gets. Inb4: "So what if I add a few spaces here and there?" - well, GPL also covers "a work based on"; this is where I (who am not a lawyer) can't speak confidently, but surely there are legal differences between "based on" and "reduced to practice", considering that both are very common occurrences in contracts, so there actually would be a lot of precedent.

No, a better subject for that analogy would be software engineers complaining about ChatGPT's coding abilities. Wagons & cars are both a matter of mechanical engineering. A digital calculator didn't displace pen-and-paper computers in the way that power drills didn't displace carpenters. Transcribing and operating a printing press are both rote procedures where the skill involved achieves accuracy and the materials involved are mostly the same. But operating Stable Diffusion and painting in PhotoShop with a stylus require wildly different modes of operation, and enjoying or being good at the latter hardly suggests that one would enjoy or be good at the former.

This sort of image generation (especially extrapolating probable upgrades & improvements of just the next few years) displaces artists without providing an upgrade path. It shouldn't take much empathy to understand how that's frustrating and scary that must be for them.

I think pxoe's expression of frustration is totally reasonable. The engineers who made this stuff could have focused on using AI to enable new possibilities, instead of undercutting existing possibilities (to create new markets instead of overtaking existing ones). They could have used 100% consensual training data, but instead felt entitled to exploit a loophole/ambiguity in the social contract under which artists have been sharing their work on the internet.

A more appropriate analogy would be, this sounds a lot like social movements complaining about capitalist co-opting of their symbols, e.g. the use of "communist" rhetoric to build oligarchies (or the sale of "save the earth" mugs made from oil-derived plastics, etc.). Even that isn't a perfect analogy, though, as the co-opted output wasn't itself the displaced work-product, although it does a better job of capturing the emotional side of it, the sense of betrayal. Ultimately, I don't think it's productive to reduce the reaction to the decisions behind Stable Diffusion etc. to a single analogy, and it shouldn't be so hard to say "sorry, you're right, this is bad for you and good for me and you have every right to express your frustration over that irreversible decision".